From Abstract to Executable Models for Multi-Agent Path Finding on Real Robots

Multi-agent path finding (MAPF) deals with the problem of finding a collision-free path for a set of agents (robots). An abstract model with a graph describing the environment and agents moving between the nodes of the graph has been proposed. This model is widely accepted by the MAPF community and majority of MAPF algorithms rely on this model. In this paper we argue that the model may not be appropriate, when the plans are to be executed on real robots. We provide some preliminary empirical evidence that abstract plans deviate from real plans executed on robots and we compare several variants of abstract models. The paper motivates further research on abstraction of problems with respect to applicability of solutions in practice.